Nonmelanoma skin cancers are the most common

Transcription

1 ORIGINAL ARTICLE See related Commentary on page iv Demarcation of Nonmelanoma Skin Cancer Margins in Thick Excisions Using Multispectral Polarized Light Imaging Anna N. Yaroslavsky, Victor Neel, and R. Rox Anderson Wellman Laboratories of Photomedicine, Department of Dermatology, Boston, Massachusetts, USA More than a million cases of nonmelanoma skin cancers are diagnosed every year. Treatment of cancer patients could bene t greatly if a real-time, noninvasive, reliable, and cost-e ective technique for delineating tumor margins were available. A novel multispectral dye-enhanced polarized light imaging technique that enables rapid imaging of large tumor elds is described. A tunable monochromatic light source and a CCD camera were employed as the imaging device. Linear polarizers were introduced into both the incident and collected light pathways in order to limit the measurement volume to the super cial tissue layers. To enhance the tumor contrast in the image, aqueous solutions of toluidine blue or methylene blue were topically applied to fresh thick skin excisions for several minutes. Then the specimens were rinsed in saline solution. Images were acquired before and after staining at the selected wavelengths. The two sets of wavelengths corresponding to the hemoglobin Soret absorption band and to the absorption bands of the dyes were used to demarcate the areas of enhanced hemoglobin and dye absorption, respectively. The resulting images demonstrate that staining signi cantly enhances contrast of the tumor in the image and enables reliable delineation of cancer. Locations and shapes of tumor lobules revealed by polarized light images closely correspond to those found in Mohs frozen sections for 41 specimens out of 45. The study demonstrates that the suggested technique has signi cant potential as a guidance tool in tumor excision surgery. Key words: dyes/multispectral polarized light imaging/nonmelanoma skin cancer/skin. J Invest Dermatol 121:259 ^266, 2003 Nonmelanoma skin cancers are the most common forms of human cancer. About 75% of all skin cancers are basal cell carcinomas (BCC) and about 20% are squamous cell carcinomas (SCC). These cancers are a major cause of morbidity in the Caucasian population. They commonly appear on sun-exposed areas of the body such as the head and neck. As many tumors occur on the face it is imperative to preserve normal skin surrounding the tumor. Unfortunately, most of these tumors have poorly de ned boundaries, which makes visual detection of the tumor borders and, consequently, precise excision a challenging problem. In the USA, Mohs micrographic surgery (MMS) (Mohs, 1941; Mikhail, 1991) is an accepted procedure that removes as little normal skin as possible while providing the highest cure rate. Using detailed mapping and complete microscopic control of the excised lesion the Mohs surgeon can pinpoint areas at the surgical margins involved with cancer that are otherwise invisible to the naked eye. Although precise and accurate, MMS is also a time-consuming and sta -intensive procedure. It requires a surgeon trained in dermatopathology, a dedicated laboratory, and a technician to prepare and evaluate frozen sections. Because of these Manuscript received January 6, 2003; revised February 24, 2003; accepted for publication March 6, 2003 Reprint requests to: Anna N. Yaroslavsky, PhD,Wellman Laboratories of Photomedicine, Department of Dermatology, 55 Fruit Street, Boston, MA yaroslav@helix.mgh.harvard.edu Abbreviations: BCC, basal cell carcinoma; CCD, charge-coupled device; DPBS, Dulbecco s phosphate-bu ered solution; MB, methylene blue; MMS, Mohs micrographic surgery; PLI, polarized light imaging; TB, toluidine blue. shortcomings, MMS is used in the minority of cases. Therefore, a simpler, more time-e cient method would be valuable. In this work, the possibility of using a multispectral dye-enhanced polarized light imaging (PLI) technique as a novel approach to the real-time noninvasive assessment of tumor margins is evaluated. Multispectral imaging is an spectroscopic technique that enables localization and e ective discrimination of the chromophores, which absorb in di erent spectral regions. This technique has been successfully used in di erent biomedical applications (Dickinson et al, 2001; Cerussi et al, 2002). Multispectral imaging may be particularly valuable for in vivo intraoperative tumor margin demarcation, as blood, which is always present in the surgical bed, makes visual (white light) inspection of the tumor margins di cult. The use of polarized light allows imaging of the super cial tissue layers only. When the light incident on the sample is linearly polarized, subtraction of two images acquired with the co-polarized (I 8 ) and cross-polarized (I > ) light can be used to largely isolate the single-scattered component, which arises mainly from super cial skin layers (Backman et al, 2000). The advantages of PLI include the ability to image comparatively thin tissue layers (E lm in the visible spectral range) and to retain a large eld of view. Jacques et al (2000) recently used white polarized light digital imaging to evaluate pigmented skin lesions. In their work a polarization image, I pol ¼ (I 8 I > )/(I 8 þ I > ), was created and analyzed. Melanin strongly scatters light (Vitkin et al, 1994), producing bright areas with excellent contrast in pigmented lesions. Such high contrast based on scattering would not be expected to occur reliably in nonmelanoma skin cancers, which contain variable amounts of melanin. Therefore, it might be bene cial to combine spectrally resolved PLI with tumor staining for enhancing tumor contrast X/03/$ Copyright r 2003 by The Society for Investigative Dermatology, Inc. 259

2 260 YAROSLAVSKY ET AL THE JOURNAL OF INVESTIGATIVE DERMATOLOGY in the images. Application of nontoxic contrast agents is practical during surgery for skin cancer. Contrast agents that alter the imaginary part of the refractive index, the dyes, increase absorption of light by the tissue. The structures that retain a dye appear darker than the surrounding medium. Dyes increase absorption of light of speci c wavelengths, thus allowing spectral selectivity. The dyes that are selectively retained by cancerous tissue have been applied previously to aid in visual examination of oral, bladder, and cervix lesions. Phenothiazinium dyes including methylene blue (MB) and toluidine blue (TB) in particular have been used for staining various carcinomas in vivo (Kaisary, 1986; Eisen et al, 1999). In addition, TB is routinely used to stain fresh-frozen tissue sections during MMS (Gross et al, 1999). Phenothiazinium dyes are accumulated to a much greater extent in mitochondria of carcinoma cells compared to normal cells (Osero et al, 1986). MB has been successfully applied to grossly demarcate neoplastic tumors in bladder (Fukui et al, 1983;Gillet al, 1984), tumors of pancreas (Fedorak et al, 1993), and Barrett s esophagus metaplasia (Canto et al, 1996). TB has been used topically to detect cervical carcinoma (Richart, 1963), oral carcinoma (Niebel and Chomet, 1965), and Barrett s esophagus metaplasia (Eisen et al, 1999). In this in vitro study the potential of the suggested multispectral dye-enhanced PLI technique to delineate the margins of nonmelanoma skin cancer was investigated. For this purpose a laboratory set-up, which enables multi-wavelength ( ve wavelengths, including 410 nm, 600 nm, 610 nm, 620 nm, and 710 nm) rapid imaging (within 10 s) of thick tissues and large surfaces (2.8 cm2.5 cm) was built. The polarized light images of thick skin excisions were acquired before and after staining with TB or MB. The acquired images were processed and super cial images for four wavelengths, including 410 nm, 600 nm, 610 nm, and 620 nm, were obtained and analyzed. The ability of the developed imaging technique to demarcate tumor margins and discriminate endogenous (blood) from exogenous (dye) chromophores was evaluated. Di erent types of BCC and SCC skin excisions were examined. The super- cial images of these lesions were compared with the corresponding Mohs frozen sections processed during Mohs surgery. Figure 1. Chemical structure of MB and TB. These two phenothiazinium dyes have similar chemical structures and exhibit similar physicochemical properties. MATERIALS AND METHODS Tissue preparation and handling Discarded tumor material was received from Mohs micrographic surgeries performed at the Dermatologic Surgery Unit of Massachusetts General Hospital under an IRB-approved protocol. In order to preserve the standard Mohs methodology, we used skin excisions that remained after Mohs histologic analysis and were therefore subject to freezing in microcryotome. After thawing in isotonic Dulbecco s phosphate-bu ered solution (DPBS, ph 7.4), the unstained tissue was imaged as described below. Then the tissue was brie y (up to 5 min) stained with 0.01%^0.05% DPBS dye solution. To remove excess of the dye after staining the specimens were rinsed in DPBS. The tissue was imaged again after staining. For imaging, the tissue was placed in a Petri dish on a gauze soaked in saline solution and covered with a coverslip. In total, 45 skin excisions of BCC (including nodular, micronodular, in ltrative, and super cial) and SCC (including invasive SCC and SCC in situ) were obtained from 38 patients. Figure 2. Absorption spectra of dermis stained with MB and/or TB. The dyes exhibit strong absorption in the 550^700 nm region. In contrast, absorption in skin, which is dominated by two main chromophores, melanin and hemoglobin, exhibits a maximum around 400 nm. Therefore spectrally resolved imaging in the range from 400 nm to 700 nm can delineate the areas of enhanced blood absorption and the areas of enhanced dye absorption. Phenothiazinium dyes Commercially available MB (MB 1% injection, USP, American Regent Laboratories, Aquadilla, Puerto Rico, USA) and TB (TB 1% AQ, LC , Fischer Scienti c Company, Pittsburgh, PA, USA) were used. MB and TB have similar chemical structures (Fig 1) and exhibit similar physicochemical properties. The blue color of the dyes is caused by the strong absorption band in the 550^700 nm region. In contrast, absorption in skin, which is dominated by two main chromophores, melanin and hemoglobin, exhibits a maximum around 400 nm. Absorption spectra of MB, TB, and human skin are presented in Fig 2. Figure 2 suggests that spectrally resolved imaging in the range from 400 nm to 700 nm could delineate the areas of enhanced blood absorption and the areas of enhanced dye absorption. Imaging equipment To enable automated multi-wavelength PLI we built a laboratory device schematically presented in Fig 3. A xenon arc lamp (Lambda LS, Sutter, Novato, CA) combined with interference lters Figure 3. Schematics of the imaging instrument.

3 VOL. 121, NO. 2 AUGUST 2003 DEMARCATION OF NONMELANOMA SKIN CANCER 261 was employed as a monochromatic light source and a CCD camera (CoolSNAP Monochrome Photometrics, Roper Scienti c, Tuscon, AZ) as an imaging device. Linearly polarizing lters were introduced into the pathways of incident light and light collected by the camera. The assembled set-up allowed acquisition and processing of conventional images and polarized light images at the selected wavelengths (k ¼410 nm, 600 nm, 610 nm, 620 nm, 710 nm) in the visible spectral range. The system provided rapid automatic image acquisition (total acquisition time for all ve wavelengths was within 10 s), a large eld of view (maximum 2.8 cm 2.5 cm), and lateral resolution of approximately 30 lm. wavelength considerably. The acquired images were processed in the following way. First, to obtain the super cial images, the di erence images I d were calculated for each wavelength. Second, to reject the background signal, the resulting images I l D ¼ Il d Ilr d,fork ¼ 410 nm, Imaging technique For the detection of skin lesions before and during surgery it is advantageous to acquire super cial images of the lesion. To achieve this goal while retaining a large eld of view we employed PLI. PLI enables super cial imaging because single scattering does not change the polarization of the elastically scattered light signi cantly, whereas polarization of the multiply scattered light is randomized. Therefore when a turbid medium such as skin is illuminated with linearly polarized light, and two images are acquired using the remitted light polarized in the directions parallel (I 8 ) and perpendicular (I > ) to the polarization of the incident light, the di erence image (I d ¼ I 8 I > ) is produced mainly by single-scattered light. The depth where the rst backscattering event occurs is an adequate approximation for the thickness of the tissue layer, which contributes dominantly to the measured signal (imaging depth). This imaging depth D is de ned by the optical properties, i.e., the scattering coe cient l s and the anisotropy factor g, of the investigated medium and can be expressed as D ¼1/l s (1 g). Using optical properties of skin from the literature (Svaasand et al, 1995;Douvenet al, 2000),itis possible to estimate the dependence of the imaging depth of the skin images on the illumination wavelength. This dependence is presented in Fig 4. It shows that the imaging depth increases with wavelength. At 400 nm the skin section thickness is approximately 75 lm, at 500 nm approximately 100 lm, and at 700 nm approximately 200 lm. As shown in Fig 2, the wavelength of 410 nm corresponds to the hemoglobin absorption band, whereas the wavelengths of 600 nm, 610 nm, and 620 nm correspond to the absorption bands of TB and MB. These wavelengths were used to demarcate the areas of enhanced hemoglobin and dye absorption, respectively. The images acquired at the reference wavelength k r ¼ 710 nm were used for the background subtraction as neither hemoglobin nor the dyes absorb the light at this Figure 4. Dependence of the imaging depth (super cial image section thickness) on the wavelength of imaging light estimated using the known optical properties of skin. Single scattering does not change polarization of the elastically scattered light, whereas polarization of the multiply scattered light is randomized. Therefore, when skin is illuminated with linearly polarized light and two images are acquired using the remitted light polarized in the directions parallel (I 8 ) and perpendicular (I > ) to the polarization of the incident light, the di erence image (I d ¼ I 8 I > )is produced mainly by single-scattered light. Thus the depth where the rst backscattering event occurs is an adequate approximation for the thickness of the tissue layer, which contributes dominantly to the measured signal (imaging depth). This imaging depth D is de ned by the scattering coe cient m s and the anisotropy factor g of the skin:d ¼1/m s (1 g). Figure 5. Images of skin with in ltrative BCC (site: lip) acquired at the wavelength l ¼ 410 nm. In the conventional images, acquired before (a) and after (b) TB staining, blood appears dark. The tumor is not apparent in the images due to high blood content and absorption (a, b) and negligible absorption of TB (b) at 410 nm. In contrast, the tumor can be clearly delineated as a structureless area in the super cial image, ID410, at 410 nm (c, arrow). Bar: 5 mm.

4 262 YAROSLAVSKY ET AL THE JOURNAL OF INVESTIGATIVE DERMATOLOGY 600 nm, 610 nm, and 620 nm, were obtained, analyzed, and compared to histopathology. Histopathology Horizontal sections were prepared by a Mohs histotechnician during surgery in the following way. Tissue removed from patients undergoing nonmelanoma skin cancer treatment was frozen in optimal cutting temperature compound and processed in the standard horizontal sectioning technique of Mohs (Mohs, 1941; Mikhail, 1991). Sections 5 lm thick were transferred to glass slides and stained with hematoxylin eosin (H&E). These frozen sections were analyzed for residual tumor at the margins. The last frozen section generated during the procedure was then compared to the super cial images of the remaining discarded piece of excision obtained using the technique described above. After imaging the tissue was xed in 10% formalin and processed for routine H&E-stained formalin- xed para n-embedded permanent histopathology for control. Horizontal sections were prepared in a way similar to the Mohs sectioning technique. RESULTS Demarcation of tumor margins and discrimination of the chromophores using multispectral dye-enhanced PLI The developed imaging technique enabled demarcation of tumor margins and discrimination of endogenous (blood) from exogenous (dye) chromophores. Example images of in ltrative morphea-form BCC obtained at the wavelengths of 410 nm and 610 nm before and after staining are presented in Figs 5 and 6. The images acquired at the wavelength of 410 nm before and after dye application are presented in Fig 5(a) and (b), respectively. In Fig 5(a) and (b) the areas contaminated with blood appear dark in the image. On the upper right and lower left boundaries of the sample the Mohs stain can be seen. Due to high blood content in the sample and enhanced hemoglobin absorption at 410 nm, it is di cult to locate the tumor in the conventional images, presented in Fig 5(a), (b). Figure 5(a) (specimen before staining) does not di er signi cantly from Fig 5(b) (specimen after staining) because, unlike blood, TB does not absorb light in the blue spectral range. Some blood was partially removed from the skin during staining and rinsing procedures, accounting for the di erences in the adherent blood patterns in the images in Fig 5(a), (b). The super cial image ID 410 at 410 nm is shown in Fig 5(c). The section thickness of the super cial image obtained at 410 nm is approximately 75 lm (see Fig 3). The absence of dark areas in the image ID 410 suggests that there is no hemoglobin within the most super cial 75 lm of the tissue. Super cial hemoglobin was removed from the skin during staining and rinsing. In Fig 6(a), (b) we present the conventional images of the same in ltrative BCC tumor acquired before (Fig 6a) and after (Fig 6b) TB staining at the wavelength of 610 nm. This wavelength corresponds to the absorption band of TB (see Fig 2). Figure 6(c) presents a super cial image, ID 610, of the tumor acquired at the same wavelength of 610 nm. In contrast to Fig 5(a), in Fig 6(a) no blood is noticeable, as at 610 nm hemoglobin absorption is weak. Comparison of Fig 6(a) (image of unstained tissue) and Fig 6(b) (image of the stained tissue) Figure 6. Images of skin with in ltrative BCC (site: lip), acquired at the wavelength k ¼ 610 nm. Comparison of the image of unstained tissue (a) and the image of the tissue stained with TB (b) shows that topical application of TB signi cantly enhances the contrast of the tumor in the image. The areas of high dye concentration appear dark in the image (b) due to enhanced absorption of TB at 610 nm. The dark area in the super cial image, ID610, of TB-stained tumor (c, arrow) compares well with H&E-stained formalin- xed para n-embedded permanent histopathology (d) and con rms that the dye stains the tumor to a much greater extent than healthy tissue. Tumor margins in (d) as determined by the Mohs surgeon are outlined with a red line. Bar: 5 mm.

5 VOL. 121, NO. 2 AUGUST 2003 DEMARCATION OF NONMELANOMA SKIN CANCER 263 con rms that the dye stains the tumor to a much greater extent than healthy tissue and shows that topical application of TB signi cantly enhances contrast of the tumor in the image. Section thickness of the super cial image at 610 nm is approximately 150 lm, which is much thicker than a Mohs frozen section (5 lm). Nonetheless, the dark area in the image clearly delineates lesion boundaries, which correlate well with the margins outlined by the surgeon in the image of the histologic slide of the same tumor (Fig 6d). It should be noted that the tumor could be identi ed in each of the super cial images presented (i.e., ID 410 (Fig 5c) andid 610 (Fig 6c)). In the images ID 410 acquired at 410 nm, the tumor appears as a structureless, homogeneous area, whereas in the image ID 610 the contrast of the tumor in comparison to normal skin is enhanced by the increased absorption of the dye that is retained in the tumor. Comparison of super cial, dye-enhanced conventional, and dye-enhanced super cial images with histopathology Example images of the nodular and micronodular BCC acquired at a wavelength of 620 nm are shown in Fig 7. Conventional and super cial images of unstained tissue are Figure 7. Images of skin with nodular and micronodular BCC (site: nose) acquired at the wavelength k ¼ 620 nm. Tumor margins are di cult to identify in the conventional image of the tissue acquired before dye application (a). In the super cial image (b) the tumor boundaries could be delineated even without staining (b, arrow). In the conventional (c) and super cial ID620 (d) images of the same specimen after TB staining the tumor is very dark and can be easily demarcated. The location and shape of the tumor in the images (b), (c), and (d) (arrows) compares well with the frozen H&E section (e) (the red line outlines the tumor margins). A detailed examination shows that in the super cial image acquired before staining and in the conventional image of the stained specimen the tumor appears as a single nest, whereas the super cial image of stained tissue reveals three closely seated tumor lobules. The frozen H&E section (e) con rms that the number and location of the tumor lobules were identi ed accurately in the image (d) and proves that PLI enables imaging of the super cial tissue layer only. Bar: 1 mm.

6 264 YAROSLAVSKY ET AL THE JOURNAL OF INVESTIGATIVE DERMATOLOGY presented in Fig 7(a) and(b), respectively. Tumor margins are hardly visible in the conventional image, whereas in the super cial image the tumor boundaries could be delineated even without staining. Conventional and super cial (ID 620 ) images of the same specimen stained with TB are shown in Fig 7(c), (d). In both images the tumor is very dark and is easily identi ed. Comparison of the images in Fig 7(b), (c), and (d) with the frozen H&E section presented in Fig 7(e) con rms that in general the location and the shape of the tumor were identi ed correctly in all the images. At the same time, the super cial image acquired before staining and the conventional image of the stained specimen make the tumorous tissue appear as a single tumor nest, whereas the super cial image of stained tissue reveals three closely seated tumor lobules. The reasons for these discrepancies are the following. The image in Fig 7(b) is formed by the light that is remitted from all the depth of the tumor sample (X 1 mm thick). Therefore the conventional image of stained tissue (Fig 7b) provides a superior contrast but does not allow any depth resolution, which makes a detailed comparison with the 5 mm thick H&E section (Fig 7e) impossible. In contrast, the super cial imaging of the unstained tumor probes the layer of 150 lm, but obviously does not provide su cient contrast to distinguish ne details of the imaged specimen. The combination of tissue staining and polarized light super cial imaging provides both strong contrast of the tumor in the image and a depth resolution of approximately 150 lm. Detailed comparison of the images in Fig 7(d) with the frozen H&E section presented in Fig 7(e) con rms that the number and location of the tumor lobules were identi ed correctly in the image ID 620 and proves that PLI enables imaging of the super cial tissue layer only. PLI examination of nonmelanoma cancers To evaluate the potential of the proposed technique we imaged several types of skin tumor excisions and compared the results to histopathology of the respective tumor. Polarized light images were diagnosed independently of the histologic images. In total, 45 skin excisions obtained from 38 patients were imaged. Thirty- ve out of 45 samples were stained with TB (dye concentration in solution varied from 0.1 to 0.5 mg per ml) and 10 samples were stained with MB (dye concentration in solution varied from 0.15 to 0.5 mg per ml). In 41 cases the location and shape of the tumors identi ed using PLI correlated well with H&E frozen sections. In three cases (two super cial BCC and one SCC in situ; all three specimens were stained using MB with dye concentrations of 0.25 mg per ml, 0.5 mg per ml, and 0.25 mg per ml, respectively) PLI located the tumors correctly, but the size of the lesions was larger in comparison to histopathology. In one case (super cial BCC, stained using TB with dye concentration of 0.1 mg per ml) PLI identi ed two more tumor nests than histopathology. In addition to the images presented and discussed above, several more example images of the di erent types of tumors, stained with TB or MB, acquired at 600 nm or 620 nm, are shown in comparison with histopathology in Figs 8^10. These are the images of in ltrative BCC, stained in TB solution with a dye concentration of 0.2 mg per ml and imaged at 600 nm (Fig 8); nodular BCC, stained in MB solution of 0.2 mg per ml and imaged at 620 nm (Fig 9); and moderately di erentiated SCC, stained in TB solution of 0.1 mg per ml and imaged at 620 nm (Fig 10). These representative images demonstrate that dyeenhanced PLI allows accurate and reliable examination of di erent types of nonmelanoma cancers. DISCUSSION The results of this in vitro study show that dye-enhanced multispectral PLI may be a suitable technique for intraoperative assessment of nonmelanoma skin tumor margins. The images Figure 8. In ltrative BCC (site: chin). Comparison of (a) the super cial image, ID600, acquired at k ¼ 600 nm (section thickness b150 lm, stain TB) with (b) a histologic frozen section prepared during Mohs surgery (section thickness b5 lm, stain H&E). Tumor margins in (b) as determined by the Mohs surgeon are outlined with a red line. General morphology of the tumor in the image compares well with that identi ed in the histopathology. Bar: 1 mm. presented in Figs 5 and 6 demonstrate how a multi-wavelength approach enables di erentiation of the tissue areas stained with blood from those stained with the dye. An e ective discrimination of the chromophores could be achieved because the absorption of the hemoglobin and the dyes (MB and TB) attain their respective maxima in di erent spectral domains (see Fig 2). A multispectral approach would be especially valuable for evaluation of specimens with high blood content, as well as for in vivo application of this technique. Staining of the tissue with MB or TB signi cantly enhanced the contrast of the tumors in the image (see Figs 5^10). Although some healthy structures, such as hair follicles, retain some dye (see conventional images of stained excisions in Figs 6 and 7), the concentration of the dye in the cancerous tissue is much higher.

7 VOL. 121, NO. 2 AUGUST 2003 DEMARCATION OF NONMELANOMA SKIN CANCER 265 Figure 10. Moderately di erentiated SCC (site: ear). Comparison of (a) the super cial image ID620 (k ¼ 620 nm, section thickness b150 lm, stain TB) and (b) a histologic frozen section prepared during Mohs surgery (section thickness b5 lm, stain H&E). Tumor margins in (b) as determined by the Mohs surgeon are outlined with a red line. General morphology of the tumor in the image compares well with that identi ed in the histopathology. Bar: 1 mm. Figure 9. Nodular BCC (site: cheek). Comparison of (a) the super cial image ID620 (k ¼ 620 nm, section thickness b150 lm, stain MB) and (b) a histologic frozen section prepared during Mohs surgery (section thickness b5 lm, stain H&E). Tumor margins in (b) as determined by the Mohs surgeon are outlined with a red line. General morphology of the tumor in the image compares well with that identi ed in the histopathology. Bar: 1 mm. Consequently, tumor-a ected areas appear much darker when imaged with light within the nm range. The results of this study suggest that all the investigated wavelengths, which are within the absorption band of TB and MB (i.e., 600 nm (Fig 8a), 610 nm (Fig 6c), and 620 nm (Fig 7d)), provide comparable and high contrast of the tumor in the image. Therefore in a clinical setting it might be possible to reduce the total acquisition time by reducing the number of wavelengths employed for imaging. As was mentioned above, the combination of tissue staining with polarized light limits the imaged volume to the super cial tissue layers and ensures that su cient contrast can be achieved for reliable di erentiation between tumor and surrounding tissue. In the wavelength range of considerable dye absorption the section thickness of skin images is approximately 150 lm, which enables a comparison of the super cial images with the histopathology (see Figs 5^10). The processed images are remarkably similar to standard MMS maps. Hair follicles, sebaceous glands, fat, and normal stromal elements are visible in detail, and with a di erent appearance from the tumor, which appears very dark due to the increased, in comparison to normal tissue, uptake of the dye. From 45 skin excisions investigated, in 41 cases PLI correlated well with H&E frozen sections. In three cases (two super cial BCC and one SCC in situ) PLI located the tumors correctly but the size of the lesions in the image was larger in comparison to histopathology. In one case (super cial BCC) PLI identi ed two more tumor nests than histopathology. There are several possible reasons for the discrepancies. First, the section thickness (or imaging depth) in PLI is in the range of 150 lm, whereas histologic H&E sections are 5 lm thick, which could explain why we see larger tumor volume in PLI in comparison to histopathology. Another possible reason could be the damage of cell membranes due to freezing/thawing artifacts that could cause leakage of the dye into the surrounding tissue. The results of this study indicate that multispectral dye-enhanced polarized light macro imaging can accurately delineate the margins of di erent types of nonmelanoma cancers, including the morphea-form BCC. The lateral resolution of our instrument is approximately 30 lm, and the section thickness of the super cial images in the red wavelength range is approximately 150 lm. A specimen 2.8 cm 2.5 cm in size can be imaged and the image processed in less than 1 min. A clinical trial with fresh specimens is under way to evaluate an optimal dye (TB versus MB), optimize the staining technique, and determine the minimal number of wavelengths required, in order to develop the multispectral PLI technique into a reliable, accurate, and timee cient method applicable for tumor demarcation in clinical

8 266 YAROSLAVSKY ET AL THE JOURNAL OF INVESTIGATIVE DERMATOLOGY practice. The rapid acquisition of these images during surgery can potentially allow tumor removal to progress without taking time to process frozen sections, the most time-consuming step in MMS. A layer of tumor can be removed, the surgical bed can be imaged, residual tumor can be detected, and guided tumor removal can take place. We thank Christina Suatengeo and John Demirs for preparation of histologic slides. REFERENCES Backman V, Wallace MB, Perelman LT, et al: Detection of preinvasive cancer cells. Nature 406:35^36, 2000 Canto MI, Setrakian S, Petras RE, Blades E, Chak A, Sivak MV Jr: Methylene blue selectively stains metaplasia in Barrett s esophagus. Gastrointest Endosc 44:1^7, 1996 Cerussi AE, Jakubowski D, Shah N, et al: Spectroscopy enhances the information content of optical mammography. JBiomedOpt7:60^71, 2002 Dickinson ME, Bearman G, Tilie S, Lansford R, Fraser SE: Multi-spectral imaging and linear unmixing add a whole new dimension to laser scanning uorescence microscopy. Biotechniques 31:1272^1278, 2001 Douven LFA, Lucassen GW: Retrieval of optical properties of skin from measurement and modelling the di use re ectance. Proc SPIE 3914:312^323, 2000 Eisen GM, Montgomery EA, Azumi N, Hartmann D-P, Bhargava P, Lippman M, Benjamin SB: Qualitative mapping of Barrett s metaplasia: A prerequisite for intervention trials. Gastrointestinal Endoscopy 50:814^818, 1999 Fedorak IJ, Ko TC, Gordon D, Flisak M, Prinz RA: Localization of islet cell tumors of pancreas: A review of current techniques. Surgery 113:242^249, 1993 Fukui I, Yokokawa M, Mitani G, et al: In vivo staining test with methylene blue for bladder cancer. JUrol130:252, 1983 Gill WB, Hu man JL, Lyon ES, Bagley DH, Schoenberg HW, Straus FH II: Selective surface staining of bladder tumors by intravesical methylene blue with enhanced endoscopic identi cation. Cancer 53:2724^2727, 1984 Gross KG, Steinman HK, Rapini RP (eds): Mohs Surgery: Fundamentals andtechniques. St Louis: Mosby, 1999 Jacques SL, Roman JR, Lee K: Imaging super cial tissues with polarized light. Las Surg Med 26:119^129, 2000 Kaisary AV: Assessment of radiotherapy in invasive bladder carcinoma using in vivo methylene blue staining technique. Urology 28:100^102, 1986 Mikhail GR (ed.): Mohs Micrographic Surgery. Philadelphia: W.B. Saunders, 1991 Mohs FE: Chemosurgery ^ a microscopically controlled method of cancer excision. Arch Surg 42:279^295, 1941 Niebel HH, Chomet B: In vivo staining test for delineation of oral intraepithelial neoplastic change: Preliminary report. JADA 68:801^805, 1965 Osero AR, Ohuoha D, Ara G, McAuli e D, Foley J, Cincotta L: Intramitochondrial dyes allow selective in vitro photolysis of carcinoma cells. Proc Natl Acad Sci USA 83:9729^9733, 1986 Richart RM: A clinical staining test for in vivo delineation of dysplasia and carcinoma in situ. Am J Obstet Gyneol. 86:703^711, 1963 Svaasand LO, Norvang LT, Fiskerstrand EJ, Stopps EKS, Berns MW, Nelson JS: Tissue parameters determining visual appearance of normal skin and port-wine stains. Las Med Sciences 10:55^65, 1995 Vitkin IA, Woolsey J, Wilson BC, Anderson RR: Optical and thermal characterization of natural (sepia o cianalis) melanin. Photochem Photobiol 59:455^462, 1994